 | A missed box near the end of a conveyor run usually sends someone back through the whole recording. The useful frame may be earlier than the visible failure, so another full watch can confirm that something went wrong without showing where it started. Grounding DINO could tag boxes in saved frames after the run and build a rough timeline. LingBot-VA 2.0 remains the action model in the attached official demo; the tagging step is a separate review idea, not part of the controller shown. The useful output is a short trail from the last normal position to the first bad action. Someone still has to check the tags before changing the controller, but at least the next review starts with a smaller piece of video. [link] [comments] |
Predictive technology: computational social science